Improve main loop pipeline

Move the new branchless core-heavy number parsing ahead of branchy dictionary lookup so that the pipeline is busy while we are searching for the slot.

Note that when using the previous branchy parser in this position the performance drops a lot, while in the previous position after the lookup the perf delta was much smaller.

Author: buybackoff

1brc/App.cs

@@ -320,10 +320,9 @@ public static unsafe void ProcessSpan(FixedDictionary<Utf8Span, Summary> result,
             while (remaining.Length > 0)
             {
                 nuint idx = remaining.IndexOfSemicolon();
-
-                result.GetValueRefOrAddDefault(new Utf8Span(remaining.Pointer, idx)).Apply(remaining.ParseIntBranchless(idx, out idx));
-
-                remaining = remaining.SliceUnsafe(idx);
+                nint value = remaining.ParseIntBranchless(idx, out var idx1);
+                result.GetValueRefOrAddDefault(new Utf8Span(remaining.Pointer, idx)).Apply(value);
+                remaining = remaining.SliceUnsafe(idx1);
             }
         }